// the find
databendlabs/databend
Data Agent Ready Warehouse : One for Analytics, Search, AI, Python Sandbox. — rebuilt from scratch. Unified architecture on your S3.
Databend is a Rust-built cloud-native OLAP warehouse that runs on object storage (S3/Azure/GCS) and unifies SQL analytics, vector search, and full-text search in one engine. It competes directly with Snowflake and ClickHouse but adds Python sandbox UDFs for running agent logic inside the warehouse. Target audience is teams building AI-heavy data pipelines who want to avoid stitching together five separate systems.
The separation of compute from storage via S3 is genuinely elastic — you spin up compute, query, tear it down, pay nothing when idle, which ClickHouse local deployments can't match. Auto schema evolution means you can ingest messy JSON without a schema ceremony upfront. The git-like data branching (time travel + snapshots) lets agents operate on production data without risking it, which is a real design win. Python sandbox UDFs running over Arrow Flight is a clever architecture — agents stay close to the data without a round-trip through your app server.
The dual Apache 2.0 + Elastic 2.0 license is a trap: enterprise features are Elastic-licensed, which means cloud providers can't host it and your legal team will have opinions before you commit to it in production. The 'agent-ready' positioning is mostly marketing wrapping around SQL + Python UDFs — there's no native tool-calling or LLM orchestration, you're assembling that yourself. The local Python driver (`databend+local:///`) is new and the maturity relative to a proper Postgres or DuckDB for local dev work is unclear. Build times in Rust for a codebase this size will punish contributors — no mention of prebuilt dev containers or sccache setup in the quick start.